!35945 runtime performance optimize-enable record actor on demand

Merge pull request !35945 from limingqi107/new_actor_runtime

mindspore/ccsrc/runtime/graph_scheduler/graph_scheduler.cc

@@ -296,7 +296,9 @@ void GraphScheduler::Clear(const ActorInfo &actor_info, const std::vector<Kernel
     for (auto &base_actor : base_actors) {
       MS_EXCEPTION_IF_NULL(base_actor);
       EraseActor(base_actor->GetAID().Name());
-      actor_manager->Terminate(base_actor->GetAID());
+      if (base_actor->parent_fusion_actor_ == nullptr) {
+        actor_manager->Terminate(base_actor->GetAID());
+      }
     }
   }
 
@@ -431,11 +433,24 @@ void GraphScheduler::BuildAndScheduleGlobalActor() {
   (void)actor_manager->Spawn(base_actor, false);
 
   // Create and schedule recorder actor.
-  auto recorder_actor = std::make_shared<RecorderActor>();
-  MS_EXCEPTION_IF_NULL(recorder_actor);
-  recorder_aid_ = &(recorder_actor->GetAID());
-  auto base_recorder_actor = static_cast<ActorReference>(recorder_actor);
-  (void)actor_manager->Spawn(base_recorder_actor, true);
+  bool recorder_actor_need = false;
+#ifndef ENABLE_SECURITY
+  if (profiler::ProfilerManager::GetInstance()->GetProfilingEnableFlag()) {
+    recorder_actor_need = true;
+  }
+#endif
+#ifdef ENABLE_DUMP_IR
+  if (mindspore::RecorderManager::Instance().RdrEnable()) {
+    recorder_actor_need = true;
+  }
+#endif
+  if (recorder_actor_need) {
+    auto recorder_actor = std::make_shared<RecorderActor>();
+    MS_EXCEPTION_IF_NULL(recorder_actor);
+    recorder_aid_ = &(recorder_actor->GetAID());
+    auto base_recorder_actor = static_cast<ActorReference>(recorder_actor);
+    (void)actor_manager->Spawn(base_recorder_actor, true);
+  }
 
   // Create and schedule debug actor.
   // debugger_actor_need is true for CPU when e2e dump is enabled and for Ascend and GPU is true when debugger or dump
@@ -548,6 +563,22 @@ void GraphScheduler::Run(ActorSet *const actor_set, const std::vector<DeviceCont
   SignalGuard sg(IntHandler);
 #endif
 
+  // Create recorder actor in the running to support the profiler in callback scene.
+#ifndef ENABLE_SECURITY
+  if (profiler::ProfilerManager::GetInstance()->GetProfilingEnableFlag() && (recorder_aid_ == nullptr)) {
+    auto recorder_actor = std::make_shared<RecorderActor>();
+    MS_EXCEPTION_IF_NULL(recorder_actor);
+    recorder_aid_ = &(recorder_actor->GetAID());
+    auto base_recorder_actor = static_cast<ActorReference>(recorder_actor);
+    auto actor_manager = ActorMgr::GetActorMgrRef();
+    MS_EXCEPTION_IF_NULL(actor_manager);
+    (void)actor_manager->Spawn(base_recorder_actor, true);
+    if (actor_set->loop_count_actor_ != nullptr) {
+      actor_set->loop_count_actor_->recorder_aid_ = recorder_aid_;
+    }
+  }
+#endif
+
   // Construct OpContext.
   OpContext<DeviceTensor> op_context;
   std::vector<Promise<int>> result(1);

tests/st/runtime/nopnode/test_runtime_optimize.py

@@ -1,50 +0,0 @@
-# Copyright 2022 Huawei Technologies Co., Ltd
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import numpy as np
-import pytest
-import mindspore
-from mindspore import context, ops, nn, Tensor
-
-
-class Net(nn.Cell):
-    def __init__(self):
-        super().__init__()
-        self.relu = ops.ReLU()
-        self.add = ops.Add()
-
-    def construct(self, input_x):
-        output = self.relu(input_x)
-        for _ in range(200):
-            output = self.add(output, 1)
-        return output
-
-
-@pytest.mark.level1
-@pytest.mark.platform_x86_cpu
-@pytest.mark.platform_x86_gpu_training
-@pytest.mark.env_onecard
-def test_multi_actor_fusion():
-    """
-    Feature: Multi actor fusion.
-    Description: Test the net which is non concurrent, that can trigger the function of multi actor fusion.
-    Expectation: The value and shape of output are the expected values.
-    """
-    context.set_context(mode=context.GRAPH_MODE)
-    x = Tensor(np.ones(2), mindspore.float32)
-    net = Net()
-    expect = np.array([201, 201])
-    for _ in range(20):
-        output = net(x)
-        assert (output.asnumpy() == expect).all()